betaFeedback

WorksPhysics and Astronomy

Detailed Design and Optimization of Ferro-Electric Tuners

Abstract

A detailed, step-by-step design methodology of a Ferroelectric Fast Reactive Tuner (FE-FRT) capable of modulating Mega VAR reactive powers on a sub-microsecond time scale is given. Closed expressions of values for all the components of the tuner are detailed, and tuner performance optimization is addressed, resulting in a Figure of Merit measure of FE-FRT tuner performance and use case applicability over a wide range of RF frequencies and reactive power levels. This enables addressing feasibility and rapid assessment of design parameters given an FE-FRT tuning scenario defined by required tuning range, cavity operating frequency a cavity stored energy.

Introduction

Frequency tuning of accelerator cavities is a fundamental tool of accelerator science and technology. The Ferroelectric Fast Reactive Tuner (FE-FRT) extends the capability of tuners towards faster and higher-power tuning for both normal conducting and superconducting cavities. This communication extends the previously developed [1, 2] theory and methodology of high-power FE-FRTs and adds a detailed, step-by-step design procedure for such an FE-FRT device and provides methods for optimizing its performance over a broad range of RF frequencies and reactive power levels. Finally, the reader is guided through the electromagnetic simulation procedure, so to permit comparison between the analytic model and electromagnetic design.

In addition to the presentation of the FE-FRT design process, one of the main conclusions of this work is the design advantage gain from FE-FRT tuner designs comprising of just two ferroelectric wafer elements. Given the challenge of designing high average power tuners, where power dissipation in the ferroelectric material is a central design driver, the interplay between power handling capability and the RF design is what is developed in this paper, in terms of FE-FRT applicability and tuner performance reach. The rationale for key design choices to determining optimum tuner parameters compliant with given frequency and tuning power requirements is given, and is followed by a comparison of analytical results with detailed numerical simulation of several challenging high power tuner scenarios. The comparisons show that analytic expressions, evaluated using Maple [3] closely approximate results obtained with CST’s Studio suites’ finite element solver [4], giving credence to the underlying design methodology.

Methodology

When tasked with designing a high power tuner, the typical inputs constraints are:

  • The cavity resonant frequency, f0f_0, ω0=2πf0\omega_0 = 2\pi f_0.

  • The required tuning range, Δf\Delta f.

  • The stored energy of the cavity, UU.

The tuner can be considered as an impedance at the port of the cavity to be tuned, with the complex power PP flowing from the cavity to the tuner. With this view, the real part of this complex power is dissipated power in the tuner, which has to be accommodated for in its design. Naturally, this dissipated power should be kept low so thermal management of the tuner is less demanding, but also to reduce undesirable draining of RF power from the cavity that would have to be replenished by the RF drive. The imaginary part of PP is the reactive power flowing into the tuner, and is responsible for the frequency tuning of the cavity. For a tuning range (lowest and highest frequency) defined by “tuner end states” and described with subscripts 1 and 2, the product of the stored energy and the required tuning range determines the reactive power that has to flow from the cavity to the tuner [1]:

S(ΔP12)=2UΔω12(1)\mathcal{S}(\Delta P_{12}) = 2U\Delta\omega_{12} \tag*{(1)}

For demanding tuner design requirements, a large reactive power is to be foreseen, and this in turn can imply additional engineering considerations to the design, in order to avoid issues such as RF voltage breakdown thresholds at the coupler port, transmission line or tuner body.

In order to assess the performance of such a reactive tuner, a Figure of Merit (FoMFoM) can be defined to quantify the ability of the tuner to produce reactive power for tuning a cavity given limits on the either power dissipation in the tuner circuit or power drawn from the cavity. The FoMFoM represents the maximal change in reactive power per the average dissipated power in the tuner system, and is a clear performance indicator for the tuner. The FoMFoM can be presented in terms of the reactance XX (used for the frequency tuning) to the resistance RR (wasted dissipation) over the full tuning range, where Z=R+jXZ = R + jX is the complex impedance of the tuner at the cavity port. The FoMFoM is given by:

FoM=ℑ(P2)−ℑ(P1)2ℜ(P1)ℜ(P2)=X2−X12R1R2(2)FoM = \frac{\Im(P_2)-\Im(P_1)}{2\sqrt{\Re(P_1)\Re(P_2)}} = \frac{X_2-X_1}{2\sqrt{R_1R_2}} \tag*{(2)}

Given that the reactance and resistance vary in a complex way as a function of the permittivity, the FoMFoM as defined is a useful but approximate indicator of performance.

Key design choices

Until now, the details of the tuner circuit were not determined, and the tuner is simply represented by a tunable impedance at the port of the cavity. Assuming a tunable reactance achieved by a ferroelectric loaded capacitor for tuning functionality, reference [1] introduced a number of potential equivalent circuit designs, with both non-resonant and resonant circuits options considered.

The Tuner as an Impedance

For achieving a high FoMFoM performance, a resonant tuner circuit with capacitive coupling to the cavity was found to be the most effective, and is taken as the starting point for this work. Figure 1 gives the general equivalent circuit, where the CfC_f, LL, RR are the resonator parameters containing the ferroelectric segment, with CSC_S is for impedance matching of tuner to the cavity. While the resonant frequency of the tuner circuit is set by the LC circuit, the choice of tuner inductance LL can be made such that the tuning range is made symmetric around f0f_0, thereby minimizing the peak voltage. Also implicit in this schematic is a transmission line, with line length set at one quarter wavelength, which transforms the tuner impedance such that at the interface to the cavity, the tuner system presents an impedance that is well defined and continuous over the full tuning range, thus symmetric around the short position on the Smith chart (see Figure 2). This configuration is appropriate for an analog tuning response.

The equivalent circuit of a capacitive coupled resonant tuner

Figure 1. The equivalent circuit of a capacitive coupled resonant tuner.

Smith chart showing an example of the tuner's impedance

Figure 2. Smith chart showing an example of the tuner’s impedance as a function of permittivity centered about a central permittivity for just the resonator (marked in red) and for the resonator followed by a quarter-wave length transmission line (marked in blue).

As previously established [1], the achievable tuning range can be derived by considering the tuner as an external impedance coupled to the cavity with transformer turn ratio given by a port impedance Z0Z_0 and a quality factor QeQ_e:

Δω12=2πΔf12=ω02QeX2−X1Z0(3)\Delta\omega_{12} = 2\pi\Delta f_{12} = \frac{\omega_0}{2Q_e}\frac{X_2-X_1}{Z_0} \tag*{(3)}

For a given tuning range, the ratio of the change in reactance X2−X1X_2-X_1 to the product QeZ0Q_eZ_0 is determined, and in what follows, the procedure for determining the required change in reactance, X2−X1X_2-X_1, that maximizes the FoMFoM is detailed.

The Ferroelectric Material

In the FE-FRT tuner the tuning function is achieved through the application of a variable bias voltage to a ferroelectric capacitor, with the applied bias voltage inducing a change in the permittivity of the ferroelectric. This tunable permittivity implies a controllable variable reactance that can be incorporated into the resonant LC circuit of the tuner, and coupled to the cavity. To realize this ferroelectric capacitance, a flat wafer format is ideal for the ferroelectric material, as it permits a reduced bias voltage for a given polarizing electric field (small wafer thickness) and improves the thermal conductivity for cooling the wafer (larger wafer surface area). In addition, reference [1] demonstrated that an annular wafer geometry reduced parasitic inductance and resistance losses at the capacitor-spacer transition. As such, the ferroelectric capacitance considered here is a stack of NwN_w of annular wafers connected in series, with the wafers separated by copper spacers that allow both for the application of bias voltage across the ferroelectric wafers and for the provision of cooling of the wafers.

Assuming an annulus wafer geometry in a multi-stack assembly of ferroelectric capacitors, the total capacitance area of the wafer stack of ferroelectric capacitors is A=Nwπ(ra2−ri2)A = N_w\pi(r_a^2-r_i^2) and is the surface area over which heat can be extracted from the ferroelectric. Here, rar_a and rir_i the outer and inner radii of the wafers, and the wafer thickness is gg.

The tuning range is mapped to a range in relative permittivity ϵ\epsilon of the ferroelectric, where ϵ\epsilon is changed by applying a bias electric field across the ferroelectric capacitor, with the full tuning range is mapped to a relative permittivity range ϵ1−ϵ2\epsilon_1-\epsilon_2 defined by the tuner end states.

In terms of the current ferroelectric of choice, a BST ceramic ((BaTiO3/SrTiO3-Mg) [5] is recommended, which has a room temperature of permittivity of ∼ϵ=160\sim\epsilon= 160, and is a low loss ceramic, with a very low frequency dependent loss tangent. The typical measured loss tangent value is δ=3.4×10−4\delta= 3.4 \times10^{-4} at 80 MHz, and a frequency scaling as ∼ω0.783\sim\omega^{0.783} between 80 MHz and 400 MHz, depending on material temperature has been reported [5].

Given that power dissipation in the ferroelectric material dominates the technical challenges when designing high-power tuners a FoMFoM for bare ferroelectric material can be derived, and as it neglects any dissipation in other elements of the tuner circuit, and this represents an upper limit on the achievable FoMFoM. This reference FoMFoM is given by

FoM(FE)=Δϵ2δϵc(4)FoM(FE) = \frac{\Delta\epsilon}{2\delta\epsilon_c} \tag*{(4)}

where Δϵ=∣ϵ2−ϵ1∣\Delta\epsilon= \lvert\epsilon_2-\epsilon_1\rvert, and ϵc\epsilon_c is a central value for the relative permittivity, approximated by ϵc≈ϵ1ϵ2\epsilon_c \approx\sqrt{\epsilon_1\epsilon_2}. For the aforementioned BST ceramic, this figure of merit is temperature dependent with a peak at about 50 °C. At this temperature, with permittivity end-states defined by a sustainable bias electric field of up to 8 MV/m, ϵ1=96\epsilon_1 = 96, ϵ2=130\epsilon_2 = 130, and the frequency dependent loss tangent at 400 MHz is δ=9.5×10−4\delta= 9.5 \times10^{-4}. This in turn implies a FoM(FE)=156FoM(FE) = 156, which can be considered an upper limit on tuner performance. For higher power tuner designs, Eq. (4) implies that a lower average permittivity enables an improved FoM(FE)FoM(FE) performance reach, but may come at the cost of limited the permittivity range, and thereby limiting tuning range.

Aspect Ratio of the Ferroelectric Capacitor

To understand the implications of power dissipation to the realization of the tuner design geometry, the power dissipated in the tuner’s circuit is distributed between the ferroelectric ceramic and the copper conductor elements and has to be considered separately. For a ferroelectric capacitor comprised of NwN_w wafers connected in series, the wafer geometry is key to optimizing the power dissipation, with the wafer area-to-thickness aspect ratio used to quantify both the ferroelectric capacitance and the thermal transport at the wafer interface. The later is relevant, as it determines the temperature rise in the wafer that must be accommodated for by a cooling system integrated into the wafer stack spacers. For example, assuming the cooling circuit provides a refrigerant at 20 °C, and the ferroelectrics operating temperature is set at 50 °C, the maximum allowable average temperature increase of the ferroelectric from dissipative losses is limited to 30 °C. This limitation is not strictly necessary but represents a compromise between maximizing the FoM(FE)FoM(FE), minimizing the wafer’s area, and addressing the engineering constraints of the cooling circuit. This is particularly relevant to the cooling circuit geometry, as for a fluid cooling system that depends on the ability to extract dissipative power from the spacers between the wafers, mechanical constraints from cooling channel dimensions must be considered; as a conservative estimate, the spacer thickness must be at least 5 mm for Nw=2N_w = 2 or 10 mm for larger NwN_w to accommodate cooling lines. It is worth noting that other methods for cooling the wafers could be envisioned, but the analytic equations provided herein are independent of the specifics of the implementation details.

To determine the optimum aspect ratio of an individual wafer, with a single-face surface area of AA, consider that tuner operation generates a power dissipation of PWP_W in the NwN_w wafer series-connected stack. Assuming that heat can be extracted from both faces of each wafer, and that the cooling can accommodate an average temperature rise ΔT\Delta T in the wafers, then for a power flow of PW/2P_W/2 flowing from the wafer center to the surface area AA

ΔT=2g∫0g/2T(y) dy=PW12KA(5)\Delta T = \frac{2}{g}\int_0^{g/2} T(y)\,dy = \frac{P_W}{12KA} \tag*{(5)}

Here KK is the thermal conductivity of the ferroelectric material [5]. Equating this ΔT\Delta T with that of the target temperature rise in the wafer relative to its cooled surface, the optimal wafer aspect ratio Aopt/gA_{opt}/g can be given by:

Aoptg=PW12ΔTK=ΔPR12NwKΔTδ(ω)ϵ1ϵ2Δϵ(6)\frac{A_{opt}}{g} = \frac{P_W}{12\Delta T K} = \frac{\Delta P_R}{12N_w K\Delta T}\frac{\delta(\omega)\sqrt{\epsilon_1\epsilon_2}}{\Delta\epsilon} \tag*{(6)}

where ΔPR=S(ΔP12)=2NwPWFoM(FE)\Delta P_R = \mathfrak{S}(\Delta P_{12}) = 2N_wP_WFoM(FE) is the peak reactive power in the complete tuner. Given that the FoMFoM improves with a reduced capacitance, this aspect ratio is considered as optimal, as the wafer stack capacitance is proportional to A/(gNW)A/(gN_W). Figure 3 shows a contour plot of this optimal aspect ratio as a function of peak reactive power in a single wafer ΔPR/Nw\Delta P_R/N_w and cavity frequency f0f_0.

Contour plot of the optimal ratio of a single wafer cross-sectional area to gap

Figure 3. The optimal ratio of a single wafer cross sectional area to gap Aopt/gA_{\mathrm{opt}}/g in meters as a function of the wafer reactive power ΔPR/Nw\Delta P_{\mathrm{R}}/N_{\mathrm{w}} in units of kVAR and operating frequency in MHz. The ferroelectric material properties are as noted in section IIB, and a wafer temperature rise of 30 degrees is assumed.

With an optimal ratio Aopt/gA_{opt}/g, the actual area and wafer thickness must be determined by engineering considerations, which are beyond the scope of this paper. However, it is noted that while the power of the bias driver of the ferroelectric capacitor varies as g2g^2 giving a preference for a smaller gg, smaller wafer areas may increase the temperature step from the spacers to the wafers, making cooling them more difficult, and suggesting a larger value of gg is preferred. As a compromise, a wafer thickness between 0.2 mm and 1 mm is recommended, depending on manufacturing feasibility.

THE RESONANT TUNER

For optimum tuner performance, the tuner equivalent circuit shown in Figure 1 should be resonant over frequencies defined by the cavity and required tuning range, and provides two important advantages over non-resonant designs:

  • The reactance change of the ferroelectric capacitor is amplified parametrically to yield a reactance change of the resonator which is QQ times larger, where QQ is the quality factor of the resonator.

  • By tuning the resonator’s frequency Ω0\Omega_0, taken at a central value of the permittivity range to about the cavity’s frequency ω0\omega_0, the impedance of the resonator is made symmetric in both end states of the tuner.

In the following sections, the constraints on the LCRLCR circuit parameters are detailed.

The Modeling the Resonator

With the LCRLCR circuit parameters as per Figure 1, CfC_f represents the ferroelectric capacitance, LL the tuner inductance including the ferroelectric capacitor and its support spacers, and R=RC+RLR = R_C + R_L, is the the total resistance of the circuit, comprised of a contribution from dielectric loss in the capacitor as well as the inductive resistance. The resonant frequency of the tuner’s circuit is then Ω=1/LC\Omega= 1/\sqrt{LC}, where CC is the total capacitance, and the resonant circuit has a shunt resistance R=QL/CR = Q\sqrt{L/C}.

The total tuner capacitance CC is the in-parallel sum

C=CsCfCs+Cf(7)C = \frac{C_s C_f}{C_s + C_f} \tag*{(7)}

and given the dissipative nature of the ferroelectric capacitor, it contributes RC=δ/(Ω0Cf)R_C = \delta/(\Omega_0 C_f) to the resistivity.

To quantify the tuner inductance LL for a stacked wafer geometry, the inductor may be considered as a short coaxial transmission line. At low frequencies (f0<30f_0 < 30 MHz) the resonator can be constructed as a compact lumped element resonant circuit, but this simplification is not relevant to most tuners, so the short terminated coaxial line modeling is adopted.

To fully model the total resistance of the resonant section, the inductor loss is taken as the series sum of the resistive loss from the ferroelectric, the resistance of the short transmission line segment, and the effect of the current at the end-wall terminations of the line representation of the inductor, so that

R=RC+RL=RC+Ri+2RB(8)R = R_C + R_L = R_C + R_i + 2R_B \tag*{(8)}

where RBR_B is the end-wall line termination resistance of the resonator. Note that the inductor is modeled as a coaxial line terminated at both ends, so RBR_B appears twice in Eq. (8).

Using the standard transmission line expression for the impedance of the coaxial stacked wafer LCRLCR resonator the resistance contribution RiR_i can be determined as

Zi=Zr02RB+Zr0tanh⁡(γl)Zr0+2RBtanh⁡(γl)∼2RB+Zr0αlr(1+tan⁡(βlr)2)+jZr0βlr=Ri+2RB+jXi(9)\begin{aligned} Z_i &= Z_{r0}\frac{2R_B + Z_{r0}\tanh(\gamma l)}{Z_{r0} + 2R_B\tanh(\gamma l)} \\ &\sim2R_B + Z_{r0}\alpha l_r(1+\tan(\beta l_r)^2) + jZ_{r0}\beta l_r \\ &= R_i + 2R_B + jX_i \tag*{(9)} \end{aligned}

where Zr0Z_{r0} is the characteristic line impedance for the resonator, γ=α+jβ\gamma= \alpha+ j\beta is the complex propagation coefficient of the line, with β=Ω/c\beta= \Omega/c, and l=lrl = l_r is the transmission line length. From this, the resistive component RiR_i is:

Ri∼Zr0αlr(1+tan⁡(βlr)2)(10)R_i \sim Z_{r0}\alpha l_r(1+\tan(\beta l_r)^2) \tag*{(10)}

where the approximation assumes the tuner is compact, and lrl_r is much shorter than the wavelength of the RF.

Here the transmission line attenuation factor α\alpha is defined for a coaxial geometry of with inner and outer conductor radii of ara_r and brb_r respectively, as

α=Rs2ηlog⁡(brar)(1ar+1br)(11)\alpha= \frac{R_s}{2\eta\log\left(\frac{b_r}{a_r}\right)}\left(\frac{1}{a_r}+\frac{1}{b_r}\right) \tag*{(11)}

where η=μ0/ϵ0=376.7Ω\eta= \sqrt{\mu_0/\epsilon_0} = 376.7\Omega is the impedance of free space, RS=πfμ0/σR_S = \sqrt{\pi f\mu_0/\sigma} is the surface resistivity, and σ\sigma is the bulk conductivity of the conductors.

Similarly, the effect of the current at the end-wall of the transmission line, represented by RBR_B, for a standard coaxial short is given by

RB=RS2πlog⁡(br/ar)(12)R_B = \frac{R_S}{2\pi}\log(b_r/a_r) \tag*{(12)}

This now permits evaluation of the impedance ZrZ_r of the tuner section, as a capacitively coupled resonator as measured at the tap point of the transmission line, and is a series-parallel combination, with

Zr=ZCS(ZCf+Zi+R)ZCS+(ZCf+Zi+R)=1jΩCS(1jΩCf+1ΩCfδ+Ri+jΩL)1jΩCS+1jΩCf+δΩCf+Ri+jΩL=−11+j2QΔΩΩ0QjΩCS[CfCS+Cf+jQ]≈−R1+j2QΔΩΩ0(CfCS+Cf)2(13)\begin{aligned} Z_r &= \frac{Z_{C_S}(Z_{C_f}+Z_i+R)}{Z_{C_S}+(Z_{C_f}+Z_i+R)} \\ &= \frac{\frac{1}{j\Omega C_S}\left(\frac{1}{j\Omega C_f}+\frac{1}{\Omega C_f}\delta+R_i+j\Omega L\right)}{\frac{1}{j\Omega C_S}+\frac{1}{j\Omega C_f}+\frac{\delta}{\Omega C_f}+R_i+j\Omega L} \\ &= -\frac{1}{1+j\frac{2Q\Delta\Omega}{\Omega_0}}\frac{Q}{j\Omega C_S}\left[\frac{C_f}{C_S+C_f}+\frac{j}{Q}\right] \\ &\approx-\frac{\mathbb{R}}{1+j\frac{2Q\Delta\Omega}{\Omega_0}}\left(\frac{C_f}{C_S+C_f}\right)^2 \tag*{(13)} \end{aligned}

Here the shunt resistance of the resonator is R=ΩLQ\mathbb{R}=\Omega LQ and ΔΩ=Ω−Ω0\Delta\Omega=\Omega-\Omega_0 is the difference in tuner resonant angular frequency compared to its average permittivity value Ω0\Omega_0 which is a function of the permittivity change (Δϵ=ϵ−ϵ0)(\Delta\epsilon=\epsilon-\epsilon_0) around the permittivity associated with Ω0\Omega_0. Balanced operational performance of the tuner implies the center frequency of the tuner’s resonator Ω0\Omega_0 is set to ω0\omega_0. The approximation in Eq. (13) is due to the strength of the quality factor QQ, and shows that matching the impedance of the resonator toward the cavity is controlled by the ratio Cf/(CS+Cf)C_f/(C_S+C_f).

Neglecting losses in the quarter-wavelength transmission line connecting the tuner to the cavity, the resonator impedance ZrZ_r is transformed by the standard transmission line equation to give the tuner load impedance ZLZ_L at the connection to the cavity port:

ZL=RL(ϵ)+jXL(ϵ)≈Z02Zr=Z021+jQCSCS+CfΔϵϵR(CfCS+Cf)2(14)\begin{aligned} Z_L &= R_L(\epsilon)+jX_L(\epsilon)\approx\frac{Z_0^2}{Z_r} \\ &= Z_0^2\frac{1+jQ\frac{C_S}{C_S+C_f}\frac{\Delta\epsilon}{\epsilon}}{\mathbb{R}\left(\frac{C_f}{C_S+C_f}\right)^2} \tag*{(14)} \end{aligned}
RL=Z02R(CS+CfCf)2R_L=\frac{Z_0^2}{\mathbb{R}}\left(\frac{C_S+C_f}{C_f}\right)^2
XL=Z02Ω0CS2CfΔϵϵX_L=Z_0^2\Omega_0\frac{C_S^2}{C_f}\frac{\Delta\epsilon}{\epsilon}

with R/Q=1/Ω0C\mathbb{R}/Q=1/\Omega_0C, and where Z0Z_0 is the characteristic impedance of the quarter-wavelength transmission line. Here the finite derivative of Ω2=1/LC\Omega^2=1/LC with respect to CC has been used to give the detuning relation

ΔΩΩ=−CSCS+CfΔϵ2ϵc(15)\frac{\Delta\Omega}{\Omega}=-\frac{C_S}{C_S+C_f}\frac{\Delta\epsilon}{2\epsilon_c} \tag*{(15)}

Given the complex tuner impedance ZL=RL+jXLZ_L=R_L+jX_L from above, and the definition of the FoM in Eq. (2), by assuming a symmetric tuning ∣ΔΩ10∣≃∣ΔΩ20∣|\Delta\Omega_{10}|\simeq|\Delta\Omega_{20}| and noting that at the extremal points (2QΔΩ/Ω0)2≫1(2Q\Delta\Omega/\Omega_0)^2\gg1, the tuner FoMFoM at the cavity can be expressed as

FoM=ΔX122R1R2≈CSCS+CfQΔϵ122ϵc=CSCfQΔϵ122ϵc(16)FoM=\frac{\Delta X_{12}}{2\sqrt{R_1R_2}}\approx\frac{C_S}{C_S+C_f}Q\frac{\Delta\epsilon_{12}}{2\epsilon_c}=\frac{C_S}{C_f}Q\frac{\Delta\epsilon_{12}}{2\epsilon_c} \tag*{(16)}

and shows that the FoMFoM depends on the relative strength of the ferroelectric capacitance CfC_f to the series capacitance CSC_S and to the quality factor QQ of the resonant tuner, but is independent of the tuner shunt impedance R\mathbb{R}.

The FoMFoM can be further simplified, as the quality factor QQ can be expressed in terms of circuit quantities; taking the resonator resistance as parameterized in terms of the ferroelectric material and coaxial modeling of the conductor losses, the quality factor of the tuner resonator segment of the tuner is

Q=1Ω0CR=1Ω0C(δΩ0Cf+Ri+2RB)(17)Q=\frac{1}{\Omega_0CR}=\frac{1}{\Omega_0C\left(\frac{\delta}{\Omega_0C_f}+R_i+2R_B\right)} \tag*{(17)}

so that the FoMFoM of Eq. (16) is independent of the series capacitance. As a cross-check, it is noted that if it is assumed the QQ of the resonator segment is solely due to the ferroelectric material loss and all conductor losses are neglected, Eq. (16) is reduced to the FoMFoM of the pure ferroelectric as given by Eq. (4).

Finally, with this circuit parametrization, the tuning range between the two permittivity endpoints can be calculated from the reactance values, and using Eqs. (3) and (14) giving

Δω12=ω02QeX1−X2Z0=ω02Z02QeCS2CfΔϵ12ϵc(18)\Delta\omega_{12}=\frac{\omega_0}{2Q_e}\frac{X_1-X_2}{Z_0}=\frac{\omega_0^2Z_0}{2Q_e}\frac{C_S^2}{C_f}\frac{\Delta\epsilon_{12}}{\epsilon_c} \tag*{(18)}

Having developed the FoMFoM parametrization in terms of equivalent circuit parameters, the circuit parameters can be chosen using procedures described in the next two sub-sections.

Determination of QeQ_e and the Series Capacitor

At this stage, the tuner has been defined and the framework for optimization established, but the coupling of the tuner to the cavity has not been addressed in detail. Optimization of this coupling is crucial for improving the tuner performance, as it impacts the choice of both the series capacitor CSC_S that sets the coupling of the resonator to the transmission line, and the external quality factor QeQ_e that sets the coupling of the full tuner system to the cavity. To address this optimization, Eq. (3) can be re-expressed by introducing the a tuner specific parameter DD

D=X2−X12Z0(19)D=\frac{X_2-X_1}{2Z_0} \tag*{(19)}

such that DD can be optimized given the reactive power, the frequency and the number of wafers of the tuner. Then, for a given DD, the required tuning range sets the value of QeQ_e, which in turn, from Eq. (18), sets the optimal capacitance CsC_s; namely

Cs=DCfΩ0Z02ϵcΔϵ21(20)C_s = \sqrt{\frac{D C_f}{\Omega_0 Z_0}\frac{2\epsilon_c}{\Delta\epsilon_{21}}} \tag*{(20)}

Again, the tuner’s resonator center frequency is required to satisfy Ω0=ω0\Omega_0 = \omega_0 ensuring the tuning range is centered around the cavity frequency. Since the ferroelectric capacitance CfC_f is determined by the geometry, and Eq. (20) sets the series capacitor CsC_s, the only adjustment left to establish the required resonance condition of the tuner is the setting of the length of the resonator lrl_r, the short coaxial segment used to model the tuner inductance. This length lrl_r is the combined length of the spacers and the wafers, and using the reactance of the tuner segment from Eq. (9) with Ω02=1/LC\Omega_0^2 = 1/LC gives lrl_r:

tan⁡(Ω0lrc)=1Ω0ZrC(21)\tan\left(\frac{\Omega_0 l_r}{c}\right) = \frac{1}{\Omega_0 Z_r C} \tag*{(21)}

where CC is the total tuner capacitance as given by Eq. (7).

Optimization of the Figure of Merit

To set the optimum tuner design, several details of the resonant tuner have not yet been specified; the first of which is the radius of the inner conductor ara_r and the related width ww of the ferroelectric annulus, w=ra−riw = r_a - r_i. These two variables are related though the area of the ferroelectric wafers:

Aopt=πNww(2ar−w)(22)A_{opt} = \pi N_w w(2a_r - w) \tag*{(22)}

such that a smaller width ww leads to a larger inner conductor radius ara_r, implying reduced dissipation in the resonator’s copper elements. The choice of actual values depends on the power, frequency and engineering constraints. It is also noted that the choice of the radii ratio b/ab/a for the transmission line has a minimal effect on the performance of the tuner and has been taken as b/a=2b/a = 2 in this work.

The second optimization is the choice of DD and brb_r, but as DD (which determines CsC_s) and the ratio θ≡br/ar\theta\equiv b_r/a_r can not be chosen independently, a simultaneous optimization of both must be made. The existence of an optimal value for θ\theta can be understood by inspecting the denominator of the tuner quality factor QQ in Eq. (17). In order to maximize QQ, Ri+2RBR_i + 2R_B must be minimized, and with explicit expressions for ZrZ_r, α\alpha and lrl_r from Eq. (21), the denominator can be re-expressed as a function of θ\theta and powers of log⁡(θ)\log(\theta). Indeed, the optimization of QQ simplifies to a minimization of the following function of θ\theta

Ri+2RB∝H1+θ−1log⁡(θ)+log⁡(θ)(23)R_i + 2R_B \propto H\frac{1+\theta^{-1}}{\log(\theta)} + \log(\theta) \tag*{(23)}

where H=H(Ω,C,ar)H = H(\Omega, C, a_r) is a tuner specific parameter

H(Ω,C,ar)≡πC2Ω02Carη(24)H(\Omega, C, a_r) \equiv\frac{\pi C}{2\Omega_0^2 C a_r \eta} \tag*{(24)}

Similarly, the existence of an optimal value of DD becomes clear from Eq. (20): Larger DD increases CsC_s and thus CC, leading to a shorter lrl_r, implying lower inductor losses. On the other hand, if the dissipation in the quarter wavelength transmission line are considered, one can show that the FoMFoM at the cavity port is maximized when DD approaches D=1D = 1. These two competing trends lead to an optimal value of DD, as shown in Figure 4b for a 1 MVAR tuner. For a 100 kVAR tuner DD stays between 1 to 3.

Given that HH has an implicit dependence on DD, it is clear that the optimization of DD and br/arb_r/a_r must be done simultaneously, and in this work, has been done numerically.

Discussion and performance curves

The tuner performance for an optimized tuner design can now be evaluated in terms of the tuner Figure of Merit (FoMFoM) as a function of cavity frequency, and power dissipated in the tuner, and Figure 5 shows the frequency dependency of the optimized FoMFoM for reactive powers of 100 kVAR and 1 MVAR, with the different curves shown corresponding to the number of stacked ferroelectric wafers. This allows the reader to quickly estimate the achievable performance of a particular tuner and match the design to the use case requirements. Given the detailed optimization discussed above, it is noted that the following design choices were made in order to generate Figure 5:

Figure of Merit of the tuner as a function of frequency for two different reactive power levels

Figure 5. The Figure of Merit of the tuner (as seen at the cavity port) as a function of frequency for two different reactive power levels. In each case the design includes optimizations described in the text.

  • The ferroelectric wafer geometry aspect ratio is adjusted as per Eq. (6), for each given frequency and power. The annulus width is set at 1 mm for the low power case and 3 mm for the high power case.

  • The series capacitor is set using to Eq. (20)

  • The resonator length lrl_r is set using Eq. (21)

  • The outer radius rfr_f of the ferroelectric is equal to the radius of the spacers ara_r

  • The quarter wavelength transmission line to the cavity has dimensions a=ara=a_r and b=2ab=2a

It is observed that the low frequency range for each value of NwN_w is truncated by a requirement that the length of the resonator lrl_r does not exceed λ/8\lambda/8, so to avoid issues with the non-linear term in the resistance RiR_i (see Eq. (10)). Similarly, the high frequency range is truncated at the point where lrl_r is less than the minimum length of the wafer spacers. In Figure 5 a 0.5 cm long spacer is assumed for the two-wafer case and 1 cm long spacers for the four and six wafer designs which limits the reactive power and hence the FoM at the higher frequency end of the curves.

In addition, the function HH has a strong dependence on frequency and tuner power (through ara_r), and for low frequency and low power designs, HH (and thus θ\theta) becomes unreasonably large. At large values of θ\theta, the frequency of the first coaxial resonator modes drop below the operating frequency which could cause unwanted mode competition. For the example shown in Figure 6 this occurs at θ≈7\theta\approx7. Consequently, the results shown in Figure 5(a) include a limit of H<3H<3, and the FoM at low frequencies is not fully optimized, as is evident from a comparison with Figure 5(b).

CST model of tuner design showing the implementation of $C_f$ and $C_s$

Figure 6. CST model of tuner design showing the implementation of CfC_f and CsC_s. Light blue represents vacuum space.

As these figures represent tuner performance over a considerable frequency range, ferroelectric wafer configurations and widely different reactive power levels, some immediate conclusions may be drawn. For tuner designs with in series stacks of ferroelectric wafer geometries, designs with more than 2 wafers does not lead to an advantage in FoM at low reactive power levels; with a 4-wafer design, there may be a small performance advantage at low frequencies, but at the price of added mechanical complication.

The FoM values, following optimization procedures, are relatively high, being of the order of 100 at 100 kVAR reactive power. At the higher reactive power of 1 MVAR, the FoM starts higher at low frequencies, even exceeding 200, but drops fast with increasing frequency.

Tuner FoM for 100 kVAR reactive power input and ferroelectric wafer 1 mm width

Figure 5a. Tuner FoM for 100 kVAR reactive power input and ferroelectric wafer 1 mm width.

Tuner FoM for 1 MVAR reactive power input and ferroelectric wafer 3 mm width

Figure 5b. Tuner FoM for 1 MVAR reactive power input and ferroelectric wafer 3 mm width.

Electromagnetic design procedure

In order to cross-check the analytical formulas from the previous sections an assessment of tuner design performance based on a detailed modelling and electromagnetic simulation has been performed, using the CST Studio Suite [4] simulation package. The tuner design example is summarized by the analytically calculated input parameters given in Table I, and represents a tuner design comprising two ferroelectric wafers of an annulus shape, capable of sustaining 1.9 MVAR reactive continuous tuning power. An outline of the design is shown in Figure 6, with the two ferroelectric wafers that comprise the capacitance CfC_f, and a (non-ferroelectric) sapphire annulus as the capacitance CsC_s shown in the insert of Figure 6. Implementing CsC_s in this way also provides an clean way to separate the resonator vacuum and the cavity vacuum. The top part of the resonator shows a break indicating that it is removable and would be sealed with a vacuum flange. It is noted that this model is not a full mechanical design of a tuner and only a representative model for electromagnetic simulation purposes.

ParameterAnalyticalUnits
f0f_0400.7948MHz
UU136J
Δf\Delta f1.1kHz
Aopt/gA_{\mathrm{opt}}/g1153.7mm
ww3mm
Z0Z_041.6Ω\Omega
aa45mm
bb90mm
ara_r62.71mm
brb_r72.77mm
QeQ_e3.25×106\times10^6

Table 1. Analytical design parameters used for comparison.

Stand Alone Tuner

To implement the detailed modelling of this high power tuner design for a 400 MHz cavity, the value of the Aopt/gA_{\mathrm{opt}}/g was set to the same value as the analytical model, and is defined by the minimum area that can sustain the power dissipated in the material. In addition, the width of the sapphire capacitor CsC_s was initially set using an electrostatic solver to estimate the capacitance. With these two values set, the model was then generated and simulated, with the tuner geometry as shown in Figure 6. The permittivity of the ferroelectric material was varied from ϵ1\epsilon_1 to ϵ2\epsilon_2, with the S11S_{11} response observed for each state. The tuner resonant frequency for each state was evaluated, and then the resonator length was set such that the the average of the two end state frequencies was at the cavity frequency, and so that the end states were equally spaced around f0f_0. A central value of the permittivity that gives this resonance at f0f_0 can be defined, as shown in Figure 7, along with the two end state resonances.

$S_{11}$ versus frequency for the two end states and a central value of $\epsilon_c$

Figure 7. S11S_{11} vs. frequency for the two end states (ϵ1\epsilon_1, ϵ2\epsilon_2), and a central value of ϵc\epsilon_c.

Attaching the Tuner to a Cavity

After the dimensions of the tuner had been approximately set, it was attached to a 2-cell, 400 MHz cavity as shown in Figure 8. The cavity had three ports, one for the FRT with QeQ_e set to the design value from Table I, a fundamental power coupler (FPC) port with Qe=5×107Q_e = 5 \times10^7 and a probe port with Qe=5×109Q_e = 5 \times10^9. The stand alone cavity with the three ports, including the FRT port was simulated separately in order to set the values of QeQ_e, where each port is terminated with its characteristic impedance. The resonant frequency of the three port cavity was also set to f0f_0, assumed to be the resonant frequency with no tuner attached. As per the analytic model, the transmission line length connecting the tuner to the cavity should be exactly λ/4\lambda/4 in length, though uncertainty in the region where the coaxial line transitions into the cavity means this is difficult to estimate. In order to properly set the transmission line length, the tuning range response as a function of line length was simulated, and the minima of the tuning range used to identify the correct transmission line length for that particular tuner configuration. Further, it was found that with the input parameters from the analytical model, the tuning range minima was only 0.97 kHz instead of the 1.1 kHz required. To increase the tuning range without affecting the other tuner parameters, QeQ_e was adjusted from 3.25×1063.25 \times10^6 to 2.95×1062.95 \times10^6, bringing the tuning range to 1.1 kHz. Finally, to have a clear comparison quantity for the CST simulation, Q‾FRT\overline{Q}_{FRT} was calculated using the known QeQ_e of the FPC and probe ports and the measured QLQ_L using:

Detail of tuner attached to two cell 400 MHz cavity

Figure 8. Detail of tuner attached to two cell 400 MHz cavity.

QFRT=11QL−1Q0−1QFPC−1Qprobe(25)Q_{FRT} = \frac{1}{\frac{1}{Q_L} - \frac{1}{Q_0} - \frac{1}{Q_{FPC}} - \frac{1}{Q_{probe}}} \tag*{(25)}

The FoMFoM was then evaluated using the average Q‾FRT\overline{Q}_{FRT} of the two end states and the tuning range, Δf\Delta f:

FoM=4πΔfQ‾FRTω0(26)FoM = \frac{4\pi\Delta f \overline{Q}_{FRT}}{\omega_0} \tag*{(26)}

and the 12% reduction in QeQ_e was expected to cause a similar reduction in the measured Q‾FRT\overline{Q}_{FRT} and hence the FOM.

Final Tuning

Final fine adjustment of parameters to center the end states around the cavity frequency f0f_0 while maintaining the required tuning range was performed using an the in-built CST optimization framework, with the following optimization objectives:

Tuning range versus transmission line length for two values of $Q_e$

Figure 9. Tuning range vs. transmission line length for two values of QeQ_e.

ParameterAnalyticalCSTUnits
Δf\Delta f1.11.1kHz
CfC_f551.9563.3pF
CsC_s578.3584pF
lrl_r18.3918.05mm
Q‾FRT\overline{Q}_{FRT}3.182.38×107\times10^7
FoMFoM81.665.3-
df1550.3554.8Hz
df2549.8550Hz
QeQ_e3.252.95×106\times10^6

Table 2. Comparison of analytical and optimized CST values for a 400 MHz tuner design.

A: Δf=∣f1−f2∣=1.1\Delta f = |f_1 - f_2| = 1.1 kHz

B: δf=∣f0−(f1+f2)/2∣<0.01\delta f = |f_0 - (f_1 + f_2)/2| < 0.01 kHz

where f1f_1 and f2f_2 are the frequencies of the two end states. Clearly, Objective A is the tuning range requirement, while Objective B enforces the tuning range to be centered on the cavity frequency with no tuner attached, f0f_0.

The inputs of the optimization were the geometry of the sapphire annulus making up CsC_s and the tuner resonator length, lrl_r, and the comparison of the analytical and CST simulation results are shown in Table II. Excellent agreement is seen between the analytical capacitance values and those estimated using an electrostatic solver. The length of the resonator also shows good agreement, indicating that the analytical model predicts the frequency of the tuner resonator well. The final S-parameters as observed through the FPC port are shown in Figure 10. It is noted that the analytical model predicts a larger value of ΔX12\Delta X_{12} than is obtained in CST for the particular tuner configuration. This implies to a need to reduce QeQ_e in the CST model, in order to increase the tuning range, and correspondingly, leads to a reduction in the CST values of both QFRTQ_{FRT} and the FoMFoM. If QeQ_e is left unchanged, QFRTQ_{FRT} can be increased to 3.1×1073.1 \times10^7, which is more in line with the analytic model, but the tuning range is reduced to 0.930.93 kHz. Finally, including all the loss mechanisms that are not included in the analytical model such as extra lengths of copper and the losses in the sapphire used to make CsC_s, further reduces QFRTQ_{FRT} by ≈4%\approx4\%.

Final S-parameters showing the tuning range

Figure 10. Final S-parameters observed through the FPC port with Qe=5×107Q_e = 5 \times10^7.

The same methodology was also applied to an 800 MHz tuner design at a power level of 458 kVAR, with the analytical values presented in Table III and comparison in Table IV. This tuner design was based on a potential use-case for a conceptual FCC-ee booster scenario whereby cavities must be detuned by 1.5 kHz every 3.8s during injection [6]. For this case QeQ_e was not adjusted to compensate for the decreased tuning range compared to the analytical estimate, resulting in a ≈15%\approx15\% reduction in tuning range. The FoMFoM is also much closer to the analytical value with the ∼13%\sim13\% difference being attributed to extra losses that are not accounted for in the analytical model. The agreement between the analytical and numeric calculations is excellent just as in the 400 MHz case.

ParameterAnalyticalUnits
f0f_0801.59MHz
UU24.3J
Δf\Delta f1.5kHz
Aopt/gA_{\mathrm{opt}}/g483.2mm
w2mm
Z0Z_041.6Ω\Omega
a32.5mm
b65mm
ara_r39.5mm
brb_r45.5mm
QeQ_e4.9×106\times10^6

Table III. Analytical design parameters (800 MHz).

ParameterAnalyticalCSTUnits
Δf\Delta f1.51.3kHz
CfC_f230.9239.5pF
CsC_s268280.6pF
lrl_r10.88.59mm
Q‾FRT\overline{Q}_{FRT}3.283.1×107\times10^7
FoMFoM56.649.3-
df1755737Hz
df2745548Hz

Table IV. Analytical vs. CST design (800 MHz).

SUMMARY

This paper presents a methodology for designing and optimizing a coaxial type resonant ferroelectric fast reactive tuner operating at high average reactive power levels over a broad range of operating frequencies. The tuner as represented by the schematic of Figure 1, comprises a ferroelectric capacitor, which has NWN_W wafers in series, separated by spacers which provide a bias voltage and cooling, A particular ferroelectric material is used in the design, with material properties taken from the ferroelectric supplier [5]. Given a cavity with a known stored energy, operating frequency and desired tuning range, the reactive power for the tuner is determined. The ferroelectric material properties together with the frequency and reactive power determine the minimum area of the ferroelectric wafers that can sustain this power. The choice of ferroelectric ring geometry and sizing then leads to a choice of the radius of the inner conductor of the tuner resonator segment, and the optimization of the FoMFoM leads to specific choices for the coupling capacitor CsC_s and the radius of the resonator’s outer conductor. Depending on the frequency and reactive power requirements, Figure 4 then gives the preferred number of ferroelectric capacitors in series NwN_w.

The optimum ratio theta as a function of H

(a) The optimum ratio θ=br/ar\theta= b_r/a_r as a function of HH

The optimum D at 1 MVAR and two wafers as a function of frequency

(b) The optimum DD at 1 MVAR and two wafers as a function of frequency

Figure 4. Optimization of the wafer aspect ratio geometry br/arb_r/a_r and operation point DD.

To illustrate the application of this design methodology to a particular case, two examples are considered, with analytic modeling of optimized tuner design compared to full electromagnetic simulation of 3-dimensional geometry. The first is a case of a very demanding tuner for a two cell 400 MHz cavity with the significant reactive power level of ΔPreactive=1.9\Delta P_{\mathrm{reactive}} = 1.9 MVAR. The second is a higher frequency (800 MHz) tuner with a reactive power level of 458 kVAR. The agreement is excellent for both examples, proving that the analytical model provided gives an accurate set of initial parameters before starting on the detailed 3D electromagnetic design.

Acknowledgments

This work is in part supported through the Innovate for Sustainable Accelerating Systems (iSAS) programme funded through the European Commission’s Horizon Europe Research and Innovation programme under Grant Agreement n°101131435.

I.B-Z. acknowledges support under the CERN Visiting Scientist program.

References

  1. [1]I. Ben-Zvi, G. Burt, A. Castilla, A. Macpherson, and N. Shipman, Conceptual design of a high reactive-power ferroelectric fast reactive tuner, Physical Review Accelerators and Beams 27, 052001 (2024).
  2. [2]I. Ben-Zvi, A. Castilla, A. Macpherson, and N. Shipman, High-power ferro-electric fast reactive tuner, arXiv preprint arXiv:2109.06806 (2021).
  3. [3]Maplesoft: Maple is a trademark of Waterloo Maple Inc., Maple (2024), https://www.maplesoft.com.
  4. [4]Dassault Systèmes, CST Studio Suite, R2024x (2024), https://www.3ds.com/products-services/simulia/products/cst-studio-suite/.
  5. [5]A. Kanareykin, E. Nenasheva, S. Karmanenko, A. Dedyk, and V. Yakovlev, Low-loss ferroelectric for accelerator applications, in Proceedings of the 2005 Particle Accelerator Conference (IEEE, 2005) pp. 4305–4307.DOI
  6. [6]F. collaboration et al., Fcc-ee: The lepton collider: Future circular collider conceptual design report volume 2, European Physical Journal Special Topics 228, 261 (2019).

Paper details

vlnc:W_v9vu8eaj
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 0

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Physics and Astronomy

Not recorded for this paper

View statement
Looking for related papers…
Contents